HS project

Combined dataset (our + Wu + Takahashi)

This file is used to combine three datasets:

• our dataset with 5 HS patients and 2 healthy donors
• Wu dataset with 6 samples from 4 healthy donors
• Takahashi dataset with 5 samples

We load each individual sample, remove melanocytes, and merge the remaining cells.

library(dplyr)
library(patchwork)
library(ggplot2)
library(ComplexHeatmap)

.libPaths()

## [1] "/usr/local/lib/R/library"

Preparation

In this section, we set the global settings of the analysis. We will store data there :

save_name = "data3"
out_dir = "."
n_threads = 5 # for tSNE

We combine the three sample information :

sample_info_1 = readRDS(paste0(out_dir, "/../1_metadata/hs_hd_sample_info.rds"))
sample_info_2 = readRDS(paste0(out_dir, "/../5_wu/1_metadata/wu_sample_info.rds"))
sample_info_3 = readRDS(paste0(out_dir, "/../6_takahashi/1_metadata/takahashi_sample_info.rds"))

column_to_keep = c("project_name", "sample_type", "sample_identifier",
                   "platform", "gender", "location", "laboratory", "color")

sample_info = rbind.data.frame(sample_info_1[, column_to_keep],
                               sample_info_2[, column_to_keep],
                               sample_info_3[, column_to_keep],
                               stringsAsFactors = FALSE)

graphics::pie(rep(1, nrow(sample_info)),
              col = sample_info$color,
              labels = sample_info$project_name)

Here are custom colors for each cell type :

color_markers = readRDS(paste0(out_dir, "/../1_metadata/hs_hd_color_markers.rds"))

data.frame(cell_type = names(color_markers),
           color = unlist(color_markers)) %>%
  ggplot2::ggplot(., aes(x = cell_type, y = 0, fill = cell_type)) +
  ggplot2::geom_point(pch = 21, size = 5) +
  ggplot2::scale_fill_manual(values = unlist(color_markers), breaks = names(color_markers)) +
  ggplot2::theme_classic() +
  ggplot2::theme(legend.position = "none",
                 axis.line = element_blank(),
                 axis.title = element_blank(),
                 axis.ticks = element_blank(),
                 axis.text.y = element_blank(),
                 axis.text.x = element_text(angle = 30, hjust = 1))

We load the markers and specific colors for each cell type :

cell_markers = readRDS(paste0(out_dir, "/../1_metadata/hs_hd_cell_markers.rds"))
lengths(cell_markers)

##          CD4 T cells          CD8 T cells     Langerhans cells 
##                   13                   13                    9 
##          macrophages              B cells              cuticle 
##                   10                   16                   15 
##               cortex              medulla                  IRS 
##                   16                   10                   16 
##        proliferative               HF-SCs            IFE basal 
##                   20                   17                   16 
## IFE granular spinous                  ORS          melanocytes 
##                   17                   15                   10 
##            sebocytes 
##                    8

We load markers to display on the dotplot :

dotplot_markers = readRDS(paste0(out_dir, "/../1_metadata/hs_hd_dotplot_markers.rds"))
dotplot_markers

## $`CD4 T cells`
## [1] "PTPRC" "CD3E"  "CD4"  
## 
## $`CD8 T cells`
## [1] "CD3E" "CD8A"
## 
## $`Langerhans cells`
## [1] "CD207" "CPVL" 
## 
## $macrophages
## [1] "TREM2" "MSR1" 
## 
## $`B cells`
## [1] "CD79A" "CD79B"
## 
## $cuticle
## [1] "MSX2"  "KRT32" "KRT35"
## 
## $cortex
## [1] "KRT31" "PRR9" 
## 
## $medulla
## [1] "BAMBI"   "ADLH1A3"
## 
## $IRS
## [1] "KRT71" "KRT73"
## 
## $proliferative
## [1] "TOP2A" "MCM5"  "TK1"  
## 
## $`HF-SCs`
## [1] "KRT14"  "CXCL14"
## 
## $`IFE basal`
## [1] "COL17A1" "KRT15"  
## 
## $`IFE granular spinous`
## [1] "SPINK5" "KRT1"  
## 
## $ORS
## [1] "KRT16" "KRT6C"
## 
## $melanocytes
## [1] "DCT"   "MLANA"
## 
## $sebocytes
## [1] "CLMP"  "PPARG"

Make data3 dataset

Individual datasets

For each sample, we :

• load individual dataset
• look at cell annotation

We load individual datasets :

sobj_list = list()

# Our data
project_names_oi = sample_info_1$project_name
sobj_list[["here"]] = lapply(project_names_oi, FUN = function(one_project_name) {
  subsobj = readRDS(paste0(out_dir, "/../2_individual/datasets/",
                           one_project_name, "_sobj_filtered.rds"))
  return(subsobj)
})
names(sobj_list[["here"]]) = project_names_oi

# Wu data
project_names_oi = sample_info_2$project_name
sobj_list[["wu"]] = lapply(project_names_oi, FUN = function(one_project_name) {
  subsobj = readRDS(paste0(out_dir, "/../5_wu/2_individual/datasets/",
                           one_project_name, "_sobj_filtered.rds"))
  return(subsobj)
})
names(sobj_list[["wu"]]) = project_names_oi

# Takahashi data
project_names_oi = sample_info_3$project_name
sobj_list[["takahashi"]] = lapply(project_names_oi, FUN = function(one_project_name) {
  subsobj = readRDS(paste0(out_dir, "/../6_takahashi/2_individual/datasets/",
                           one_project_name, "_sobj_filtered.rds"))
  return(subsobj)
})
names(sobj_list[["takahashi"]]) = project_names_oi

# Unlist
sobj_list = unlist(sobj_list, recursive = FALSE)

lapply(sobj_list, FUN = dim) %>%
  do.call(rbind, .) %>%
  rbind(., colSums(.))

##                        [,1]  [,2]
## here.2021_31          27955  1043
## here.2021_36          27955   602
## here.2021_41          27955  2256
## here.2022_03          27955  3977
## here.2022_14          27955  2588
## here.2022_01          27955  1213
## here.2022_02          27955  2286
## wu.F18                27955  1372
## wu.F31B               27955  4786
## wu.F31W               27955  3520
## wu.F59                27955  2445
## wu.F62B               27955  3279
## wu.F62W               27955  2360
## takahashi.GSM3717034  10320    71
## takahashi.GSM3717035  12129   275
## takahashi.GSM3717036  14170   510
## takahashi.GSM3717037  32458  4084
## takahashi.GSM3717038  32458  1094
##                      464950 37761

We represent cells in the tSNE :

name2D = "RNA_pca_20_tsne"

We look at cell type annotation for each dataset :

plot_list = lapply(sobj_list, FUN = function(one_sobj) {
  mytitle = as.character(unique(one_sobj$project_name))
  mysubtitle = ncol(one_sobj)
  
  if (!(name2D %in% names(one_sobj@reductions))) {
    name2D = names(one_sobj@reductions)[2]
  }
  
  p = Seurat::DimPlot(one_sobj, group.by = "cell_type",
                      reduction = name2D) +
    ggplot2::scale_color_manual(values = color_markers,
                                breaks = names(color_markers),
                                name = "Cell Type") +
    ggplot2::labs(title = mytitle,
                  subtitle = paste0(mysubtitle, " cells")) +
    ggplot2::theme(aspect.ratio = 1,
                   plot.title = element_text(hjust = 0.5),
                   plot.subtitle = element_text(hjust = 0.5)) +
    Seurat::NoAxes()
  
  return(p)
})

plot_list[[length(plot_list) + 1]] = patchwork::guide_area()

patchwork::wrap_plots(plot_list, ncol = 4) +
  patchwork::plot_layout(guides = "collect") &
  ggplot2::theme(legend.position = "right")

and clustering :

plot_list = lapply(sobj_list, FUN = function(one_sobj) {
  mytitle = as.character(unique(one_sobj$project_name))
  mysubtitle = ncol(one_sobj)
  
  if (!(name2D %in% names(one_sobj@reductions))) {
    name2D = names(one_sobj@reductions)[2]
  }
  
  p = Seurat::DimPlot(one_sobj, group.by = "seurat_clusters",
                      reduction = name2D, label = TRUE) +
    ggplot2::labs(title = mytitle,
                  subtitle = paste0(mysubtitle, " cells")) +
    ggplot2::theme(aspect.ratio = 1,
                   plot.title = element_text(hjust = 0.5),
                   plot.subtitle = element_text(hjust = 0.5)) +
    Seurat::NoAxes() + Seurat::NoLegend()
  
  return(p)
})

patchwork::wrap_plots(plot_list, ncol = 4)

Melanocytes removal

For each individual dataset, we remove melanocytes. First, we smooth cell type annotation at a cluster level :

sobj_list = lapply(sobj_list, FUN = function(one_sobj) {
  cluster_type = table(one_sobj$cell_type, one_sobj$seurat_clusters) %>%
    prop.table(., margin = 2) %>%
    apply(., 2, which.max)
  cluster_type = setNames(nm = names(cluster_type),
                          levels(one_sobj$cell_type)[cluster_type])
  
  one_sobj$cluster_type = cluster_type[one_sobj$seurat_clusters]
  
  ## Output
  return(one_sobj)
})

To locate melanocytes, we look at their score, cell type annotation, and clustering.

plot_list = lapply(sobj_list, FUN = function(one_sobj) {
  project_name = as.character(unique(one_sobj$project_name))
  plot_sublist = list()
  
  if (!(name2D %in% names(one_sobj@reductions))) {
    name2D = names(one_sobj@reductions)[2]
  }
  
  # Score
  plot_sublist[[1]] = Seurat::FeaturePlot(one_sobj, reduction = name2D,
                                          features = "score_melanocytes") +
    ggplot2::labs(title = project_name,
                  subtitle = "Melanocytes score") +
    Seurat::NoAxes() +
    ggplot2::scale_color_gradientn(colors = aquarius:::color_gene) +
    ggplot2::theme(aspect.ratio = 1,
                   plot.subtitle = element_text(hjust = 0.5))
  
  # Cell type
  plot_sublist[[2]] = Seurat::DimPlot(one_sobj,
                                      reduction = name2D,
                                      group.by = "cell_type",
                                      order = "melanocytes") +
    ggplot2::scale_color_manual(values = c("purple", rep("gray92", length(color_markers) - 1)),
                                breaks = c("melanocytes", setdiff(names(color_markers), "melanocytes"))) +
    ggplot2::labs(title = "Cell type annotation",
                  subtitle = paste0(sum(one_sobj$cell_type == "melanocytes"),
                                    " melanocytes")) +
    Seurat::NoAxes() + Seurat::NoLegend() +
    ggplot2::theme(aspect.ratio = 1,
                   plot.title = element_text(hjust = 0.5),
                   plot.subtitle = element_text(hjust = 0.5))
  
  # Clusters
  plot_sublist[[3]] = Seurat::DimPlot(one_sobj,
                                      reduction = name2D,
                                      group.by = "seurat_clusters",
                                      label = TRUE) +
    ggplot2::labs(title = "Clusters") +
    Seurat::NoAxes() + Seurat::NoLegend() +
    ggplot2::theme(aspect.ratio = 1,
                   plot.title = element_text(hjust = 0.5),
                   plot.subtitle = element_text(hjust = 0.5))
  
  # Cluster type
  plot_sublist[[4]] = Seurat::DimPlot(one_sobj,
                                      reduction = name2D,
                                      group.by = "cluster_type") +
    ggplot2::scale_color_manual(values = c("purple", rep("gray92", length(color_markers) - 1)),
                                breaks = c("melanocytes", setdiff(names(color_markers), "melanocytes"))) +
    ggplot2::labs(title = "Cluster annotation",
                  subtitle = paste0(sum(one_sobj$cluster_type == "melanocytes"),
                                    " melanocytes")) +
    Seurat::NoAxes() + Seurat::NoLegend() +
    ggplot2::theme(aspect.ratio = 1,
                   plot.title = element_text(hjust = 0.5),
                   plot.subtitle = element_text(hjust = 0.5))
  
  return(plot_sublist)
}) %>% unlist(., recursive = FALSE)

patchwork::wrap_plots(plot_list, ncol = 4)

We remove melanocytes based on cluster annotation for 10X datasets and based on the cell type annotation for Drop-Seq datasets :

sobj_list = lapply(sobj_list, FUN = function(one_sobj) {
  if (one_sobj@project.name %in% c("GSM3717034", "GSM3717035", "GSM3717036")) {
    one_sobj$is_of_interest = (one_sobj$cell_type != "melanocytes")
  } else {
    one_sobj$is_of_interest = (one_sobj$cluster_type != "melanocytes")
  }
  
  if (sum(one_sobj$is_of_interest) > 0) {
    one_sobj = subset(one_sobj, is_of_interest == TRUE)
  } else {
    one_sobj = NA
  }
  
  one_sobj$is_of_interest = NULL
  return(one_sobj)
})

lapply(sobj_list, FUN = dim) %>%
  do.call(rbind, .) %>%
  rbind(., colSums(.))

##                        [,1]  [,2]
## here.2021_31          27955   885
## here.2021_36          27955   528
## here.2021_41          27955  1982
## here.2022_03          27955  3457
## here.2022_14          27955  2422
## here.2022_01          27955   835
## here.2022_02          27955  2002
## wu.F18                27955  1372
## wu.F31B               27955  4624
## wu.F31W               27955  3520
## wu.F59                27955  2445
## wu.F62B               27955  3221
## wu.F62W               27955  2360
## takahashi.GSM3717034  10320    65
## takahashi.GSM3717035  12129   270
## takahashi.GSM3717036  14170   497
## takahashi.GSM3717037  32458  3781
## takahashi.GSM3717038  32458  1023
##                      464950 35289

Re-annotation

We remove melanocytes from annotation :

cell_markers = cell_markers[names(cell_markers) != "melanocytes"]
color_markers = color_markers[names(color_markers) != "melanocytes"]
dotplot_markers = dotplot_markers[names(dotplot_markers) != "melanocytes"]

We re-annotate cells for cell type, since melanocytes have been removed :

sobj_list = lapply(sobj_list, FUN = function(one_sobj) {
  # Remove old annotation
  one_sobj@meta.data[, grep(colnames(one_sobj@meta.data), pattern = "score", value = TRUE)] = NULL
  
  # Re-annot
  one_sobj = aquarius::cell_annot_custom(one_sobj,
                                         newname = "cell_type",
                                         markers = cell_markers,
                                         use_negative = TRUE,
                                         add_score = FALSE,
                                         verbose = TRUE)
  
  # Set factor levels
  one_sobj$cell_type = factor(one_sobj$cell_type, levels = names(cell_markers))
  
  return(one_sobj)
})

Gene annotation

Our dataset and Wu dataset were processed using the same annotation. In Takahashi dataset, all genes are not shared across datasets:

Note: With the ggvenn package, this is not possible to make a Venn diagram with 5 sets.

ggvenn::ggvenn(data = list(
  here.2021_31 = sobj_list[["here.2021_31"]]@assays[["RNA"]]@meta.features$Ensembl_ID,
  wu.F18 = sobj_list[["wu.F18"]]@assays[["RNA"]]@meta.features$Ensembl_ID,
  takahashi.GSM3717034 = sobj_list[["takahashi.GSM3717034"]]@assays[["RNA"]]@meta.features$Ensembl_ID,
  takahashi.GSM3717038 = sobj_list[["takahashi.GSM3717038"]]@assays[["RNA"]]@meta.features$Ensembl_ID),
  stroke_size = 0.5, set_name_size = 4) +
  ggplot2::labs(title = "Gene Ensembl IDs between the 4 datasets") +
  ggplot2::theme(plot.title = element_text(hjust = 0.5, face = "bold"))

We keep common genes between all datasets + common genes between the 10X datasets, based on the EnsemblID

# All Ensembl IDs
common_genes = lapply(sobj_list, FUN = function(one_sobj) {
  ensembl_id = one_sobj@assays[["RNA"]]@meta.features$Ensembl_ID
  
  return(ensembl_id)
})
names(common_genes) = names(sobj_list)

# Common between 10X datasets
common_genes_10x = common_genes[!(names(common_genes) %in% c("takahashi.GSM3717034",
                                                             "takahashi.GSM3717035",
                                                             "takahashi.GSM3717036"))] %>%
  Reduce(intersect, .)

# Common between all
common_genes = Reduce(intersect, common_genes)

# Venn diagram
ggvenn::ggvenn(data = list(
  common_all = common_genes,
  common_10X = common_genes_10x),
  stroke_size = 0.5, set_name_size = 4) +
  ggplot2::labs(title = "Gene Ensembl IDs") +
  ggplot2::theme(plot.title = element_text(hjust = 0.5, face = "bold"))

We keep the union of all these genes :

common_genes = union(common_genes, common_genes_10x)
rm(common_genes_10x)

length(common_genes)

## [1] 25605

To which gene names they correspond, in one of our dataset ?

gene_corresp = sobj_list[["here.2021_31"]]@assays$RNA@meta.features %>%
  dplyr::filter(Ensembl_ID %in% common_genes) %>%
  dplyr::select(Ensembl_ID, gene_name)

dim(gene_corresp)

## [1] 25605     2

head(gene_corresp)

##                  Ensembl_ID   gene_name
## MIR1302-2HG ENSG00000243485 MIR1302-2HG
## FAM138A     ENSG00000237613     FAM138A
## OR4F5       ENSG00000186092       OR4F5
## AL627309.1  ENSG00000238009  AL627309.1
## AL627309.3  ENSG00000239945  AL627309.3
## AL627309.4  ENSG00000241599  AL627309.4

We subset Seurat object for the Ensembl IDs of interest.

sobj_list = lapply(sobj_list, FUN = function(one_sobj) {
  # Extract metadata
  one_metadata = one_sobj@meta.data
  
  # Extract and subset gene annotation
  one_annotation = one_sobj@assays[["RNA"]]@meta.features %>%
    dplyr::filter(Ensembl_ID %in% gene_corresp$Ensembl_ID)
  
  # Subset gene corresp for reordering
  one_gene_corresp = gene_corresp %>%
    dplyr::filter(Ensembl_ID %in% one_annotation$Ensembl_ID)
  
  # Extract count matrix and subset genes
  one_count_matrix = one_sobj@assays[["RNA"]]@counts
  one_count_matrix = one_count_matrix[rownames(one_annotation), ]
  
  # Reorder according to the gene correspondence
  gene_order = match(one_gene_corresp$Ensembl_ID,
                     one_annotation$Ensembl_ID)
  
  # Reorder the count matrix and annotation
  one_annotation = one_annotation[gene_order, ]
  one_count_matrix = one_count_matrix[gene_order, ]
  rownames(one_count_matrix) = rownames(one_gene_corresp)
  
  # Build again the Seurat object
  one_sobj = Seurat::CreateSeuratObject(counts = one_count_matrix,
                                        meta.data = one_metadata)
  one_sobj@assays[["RNA"]]@meta.features = one_gene_corresp
  
  return(one_sobj)
})

lapply(sobj_list, FUN = dim) %>%
  do.call(rbind, .) %>%
  rbind(., colSums(.))

##                        [,1]  [,2]
## here.2021_31          25605   885
## here.2021_36          25605   528
## here.2021_41          25605  1982
## here.2022_03          25605  3457
## here.2022_14          25605  2422
## here.2022_01          25605   835
## here.2022_02          25605  2002
## wu.F18                25605  1372
## wu.F31B               25605  4624
## wu.F31W               25605  3520
## wu.F59                25605  2445
## wu.F62B               25605  3221
## wu.F62W               25605  2360
## takahashi.GSM3717034   8926    65
## takahashi.GSM3717035  10676   270
## takahashi.GSM3717036  11842   497
## takahashi.GSM3717037  25605  3781
## takahashi.GSM3717038  25605  1023
##                      415519 35289

Combined dataset

We combine all datasets :

sobj = base::merge(sobj_list[[1]],
                   y = sobj_list[c(2:length(sobj_list))],
                   add.cell.ids = names(sobj_list))
sobj

## An object of class Seurat 
## 25605 features across 35289 samples within 1 assay 
## Active assay: RNA (25605 features, 0 variable features)

We add again the correspondence between gene names and gene ID. We take the correspondence from one individual 10X dataset.

sobj@assays$RNA@meta.features = sobj_list[[1]]@assays$RNA@meta.features[, c("Ensembl_ID", "gene_name")]

head(sobj@assays$RNA@meta.features)

##                  Ensembl_ID   gene_name
## MIR1302-2HG ENSG00000243485 MIR1302-2HG
## FAM138A     ENSG00000237613     FAM138A
## OR4F5       ENSG00000186092       OR4F5
## AL627309.1  ENSG00000238009  AL627309.1
## AL627309.3  ENSG00000239945  AL627309.3
## AL627309.4  ENSG00000241599  AL627309.4

We remove the list of objects :

rm(sobj_list)

We keep a subset of meta.data and reset levels :

sobj@meta.data = sobj@meta.data[, c("orig.ident", "nCount_RNA", "nFeature_RNA", "log_nCount_RNA",
                                    "project_name", "sample_identifier", "sample_type",
                                    "laboratory", "location", "Seurat.Phase", "cyclone.Phase",
                                    "percent.mt", "percent.rb", "cell_type")]

sobj$orig.ident = factor(sobj$orig.ident, levels = levels(sample_info$project_name))
sobj$project_name = factor(sobj$project_name, levels = levels(sample_info$project_name))
sobj$sample_identifier = factor(sobj$sample_identifier, levels = levels(sample_info$sample_identifier))
sobj$sample_type = factor(sobj$sample_type, levels = unique(sample_info$sample_type))
sobj$cell_type = factor(sobj$cell_type, levels = names(color_markers))

summary(sobj@meta.data)

##       orig.ident      nCount_RNA      nFeature_RNA  log_nCount_RNA  
##  F31B      : 4624   Min.   :   318   Min.   : 250   Min.   : 5.762  
##  GSM3717037: 3781   1st Qu.:  3274   1st Qu.:1226   1st Qu.: 8.094  
##  F31W      : 3520   Median :  8707   Median :2298   Median : 9.072  
##  2022_03   : 3457   Mean   : 11627   Mean   :2438   Mean   : 8.877  
##  F62B      : 3221   3rd Qu.: 16354   3rd Qu.:3371   3rd Qu.: 9.702  
##  F59       : 2445   Max.   :139803   Max.   :7942   Max.   :11.848  
##  (Other)   :14241                                                   
##      project_name        sample_identifier sample_type  laboratory       
##  F31B      : 4624   Wu_HD_2       : 4624   HS: 9274    Length:35289      
##  GSM3717037: 3781   Takahashi_HD_4: 3781   HD:26015    Class :character  
##  F31W      : 3520   Wu_HD_3       : 3520               Mode  :character  
##  2022_03   : 3457   HS_4          : 3457                                 
##  F62B      : 3221   Wu_HD_5       : 3221                                 
##  F59       : 2445   Wu_HD_4       : 2445                                 
##  (Other)   :14241   (Other)       :14241                                 
##    location         Seurat.Phase       cyclone.Phase        percent.mt    
##  Length:35289       Length:35289       Length:35289       Min.   : 0.000  
##  Class :character   Class :character   Class :character   1st Qu.: 2.778  
##  Mode  :character   Mode  :character   Mode  :character   Median : 4.516  
##                                                           Mean   : 5.360  
##                                                           3rd Qu.: 6.868  
##                                                           Max.   :20.000  
##                                                                           
##    percent.rb                     cell_type   
##  Min.   : 0.4948   ORS                 :9121  
##  1st Qu.:19.1530   IFE granular spinous:5349  
##  Median :23.9507   IFE basal           :4117  
##  Mean   :23.2444   HF-SCs              :3292  
##  3rd Qu.:28.2717   cuticle             :3021  
##  Max.   :48.0392   proliferative       :2590  
##                    (Other)             :7799

Processing

We remove genes that are expressed in less than 5 cells :

sobj = aquarius::filter_features(sobj, min_cells = 5)

## [1] 25605 35289
## [1] 19849 35289

sobj

## An object of class Seurat 
## 19849 features across 35289 samples within 1 assay 
## Active assay: RNA (19849 features, 0 variable features)

Metadata

How many cells by sample ?

table(sobj$project_name)

## 
##    2021_31    2021_36    2021_41    2022_03    2022_14    2022_01    2022_02 
##        885        528       1982       3457       2422        835       2002 
##        F18       F31B       F31W        F59       F62B       F62W GSM3717034 
##       1372       4624       3520       2445       3221       2360         65 
## GSM3717035 GSM3717036 GSM3717037 GSM3717038 
##        270        497       3781       1023

We represent this information as a barplot :

aquarius::plot_barplot(df = table(sobj$project_name,
                                  sobj$cell_type) %>%
                         as.data.frame.table() %>%
                         `colnames<-`(c("Sample", "Cell Type", "Number")),
                       x = "Sample", y = "Number", fill = "Cell Type",
                       position = position_fill()) +
  ggplot2::scale_fill_manual(values = unlist(color_markers),
                             breaks = names(color_markers),
                             name = "Cell Type")

This is the same barplot with another position :

aquarius::plot_barplot(df = table(sobj$project_name,
                                  sobj$cell_type) %>%
                         as.data.frame.table() %>%
                         `colnames<-`(c("Sample", "Cell Type", "Number")),
                       x = "Sample", y = "Number", fill = "Cell Type",
                       position = position_stack()) +
  ggplot2::scale_fill_manual(values = unlist(color_markers),
                             breaks = names(color_markers),
                             name = "Cell Type")

Projection

We normalize the count matrix for remaining cells and select highly variable features :

sobj = Seurat::NormalizeData(sobj,
                             normalization.method = "LogNormalize")
sobj = Seurat::FindVariableFeatures(sobj, nfeatures = 2000)
sobj = Seurat::ScaleData(sobj)

sobj

## An object of class Seurat 
## 19849 features across 35289 samples within 1 assay 
## Active assay: RNA (19849 features, 2000 variable features)

We perform a PCA :

sobj = Seurat::RunPCA(sobj,
                      assay = "RNA",
                      reduction.name = "RNA_pca",
                      npcs = 100,
                      seed.use = 1337L)
sobj

## An object of class Seurat 
## 19849 features across 35289 samples within 1 assay 
## Active assay: RNA (19849 features, 2000 variable features)
##  1 dimensional reduction calculated: RNA_pca

We choose the number of dimensions such that they summarize 60 % of the variability :

stdev = sobj@reductions[["RNA_pca"]]@stdev
stdev_prop = cumsum(stdev)/sum(stdev)
ndims = which(stdev_prop > 0.60)[1]
ndims

## [1] 37

We can visualize this on the elbow plot :

elbow_p = Seurat::ElbowPlot(sobj, ndims = 100, reduction = "RNA_pca") +
  ggplot2::geom_point(x = ndims, y = stdev[ndims], col = "red")
x_text = ggplot_build(elbow_p)$layout$panel_params[[1]]$x$get_labels() %>% as.numeric()
elbow_p = elbow_p +
  ggplot2::scale_x_continuous(breaks = sort(c(x_text, ndims)), limits = c(0, 100))
x_color = ifelse(ggplot_build(elbow_p)$layout$panel_params[[1]]$x$get_labels() %>%
                   as.numeric() %>% round(., 2) == round(ndims, 2), "red", "black")
elbow_p = elbow_p +
  ggplot2::theme_classic() +
  ggplot2::theme(axis.text.x = element_text(color = x_color))

elbow_p

We generate a tSNE and a UMAP with 37 principal components :

sobj = Seurat::RunTSNE(sobj,
                       reduction = "RNA_pca",
                       dims = 1:ndims,
                       seed.use = 1337L,
                       num_threads = n_threads, # Rtsne::Rtsne option
                       reduction.name = paste0("RNA_pca_", ndims, "_tsne"))

sobj = Seurat::RunUMAP(sobj,
                       reduction = "RNA_pca",
                       dims = 1:ndims,
                       seed.use = 1337L,
                       reduction.name = paste0("RNA_pca_", ndims, "_umap"))

(Time to run : 124.05 s)

We can visualize the two representations :

tsne = Seurat::DimPlot(sobj, group.by = "project_name",
                       reduction = paste0("RNA_pca_", ndims, "_tsne")) +
  ggplot2::scale_color_manual(values = sample_info$color,
                              breaks = sample_info$project_name) +
  Seurat::NoAxes() + ggplot2::ggtitle("PCA - tSNE") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5),
                 legend.position = "none")

umap = Seurat::DimPlot(sobj, group.by = "project_name",
                       reduction = paste0("RNA_pca_", ndims, "_umap")) +
  ggplot2::scale_color_manual(values = sample_info$color,
                              breaks = sample_info$project_name) +
  Seurat::NoAxes() + ggplot2::ggtitle("PCA - UMAP") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))

tsne | umap

Batch-effect correction

We remove sample specific effect on the pca using Harmony :

`%||%` = function(lhs, rhs) {
  if (!is.null(x = lhs)) {
    return(lhs)
  } else {
    return(rhs)
  }
}

set.seed(1337L)
sobj = harmony::RunHarmony(object = sobj,
                           group.by.vars = "project_name",
                           plot_convergence = TRUE,
                           reduction = "RNA_pca",
                           assay.use = "RNA",
                           reduction.save = "harmony",
                           max.iter.harmony = 50,
                           project.dim = FALSE)

(Time to run : 198.87 s)

From this batch-effect removed projection, we generate a tSNE and a UMAP.

sobj = Seurat::RunUMAP(sobj, 
                       seed.use = 1337L,
                       dims = 1:ndims,
                       reduction = "harmony",
                       reduction.name = paste0("harmony_", ndims, "_umap"),
                       reduction.key = paste0("harmony_", ndims, "umap_"))

sobj = Seurat::RunTSNE(sobj,
                       dims = 1:ndims,
                       seed.use = 1337L,
                       num_threads = n_threads, # Rtsne::Rtsne option
                       reduction = "harmony",
                       reduction.name = paste0("harmony_", ndims, "_tsne"),
                       reduction.key = paste0("harmony", ndims, "tsne_"))

(Time to run : 127.66 s)

We visualize the corrected projections :

tsne = Seurat::DimPlot(sobj, group.by = "project_name",
                       reduction = paste0("harmony_", ndims, "_tsne")) +
  ggplot2::scale_color_manual(values = sample_info$color,
                              breaks = sample_info$project_name) +
  Seurat::NoAxes() + ggplot2::ggtitle("PCA - harmony - tSNE") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5),
                 legend.position = "none")

umap = Seurat::DimPlot(sobj, group.by = "project_name",
                       reduction = paste0("harmony_", ndims, "_umap")) +
  ggplot2::scale_color_manual(values = sample_info$color,
                              breaks = sample_info$project_name) +
  Seurat::NoAxes() + ggplot2::ggtitle("PCA - harmony - UMAP") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))

tsne | umap

We will keep the tSNE from harmony :

reduction = "harmony"
name2D = paste0("harmony_", ndims, "_tsne")

Clustering

We generate a clustering :

sobj = Seurat::FindNeighbors(sobj, reduction = reduction, dims = 1:ndims)
sobj = Seurat::FindClusters(sobj, resolution = 1.2)

## Modularity Optimizer version 1.3.0 by Ludo Waltman and Nees Jan van Eck
## 
## Number of nodes: 35289
## Number of edges: 1441587
## 
## Running Louvain algorithm...
## Maximum modularity in 10 random starts: 0.8872
## Number of communities: 33
## Elapsed time: 8 seconds

clusters_plot = Seurat::DimPlot(sobj, reduction = name2D, label = TRUE) +
  Seurat::NoAxes() + Seurat::NoLegend() +
  ggplot2::labs(title = "Clusters ID") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))
clusters_plot

Visualization

We represent the 4 quality metrics :

plot_list = Seurat::FeaturePlot(sobj, reduction = name2D,
                                combine = FALSE, pt.size = 0.25,
                                features = c("percent.mt", "percent.rb", "log_nCount_RNA", "nFeature_RNA"))
plot_list = lapply(plot_list, FUN = function(one_plot) {
  one_plot +
    Seurat::NoAxes() +
    ggplot2::scale_color_gradientn(colors = aquarius:::color_gene) +
    ggplot2::theme(aspect.ratio = 1)
})

patchwork::wrap_plots(plot_list, nrow = 1)

Cell type

We visualize cell type :

plot_list = lapply((c(paste0("RNA_pca_", ndims, "_tsne"),
                      paste0("RNA_pca_", ndims, "_umap"),
                      paste0("harmony_", ndims, "_tsne"),
                      paste0("harmony_", ndims, "_umap"))),
                   FUN = function(one_red) {
                     Seurat::DimPlot(sobj, group.by = "cell_type",
                                     reduction = one_red,
                                     cols = color_markers) +
                       Seurat::NoAxes() + ggplot2::ggtitle(one_red) +
                       ggplot2::theme(aspect.ratio = 1,
                                      plot.title = element_text(hjust = 0.5))
                   })

patchwork::wrap_plots(plot_list, nrow = 2) +
  patchwork::plot_layout(guides = "collect")

We make a representation split by origin to show cell types :

plot_list = aquarius::plot_split_dimred(sobj,
                                        reduction = name2D,
                                        split_by = "project_name",
                                        split_color = setNames(sample_info$color,
                                                               nm = sample_info$project_name),
                                        group_by = "cell_type",
                                        group_color = color_markers)

plot_list[[length(plot_list) + 1]] = patchwork::guide_area()

patchwork::wrap_plots(plot_list, ncol = 4) +
  patchwork::plot_layout(guides = "collect") &
  ggplot2::theme(legend.position = "right")

Laboratory

We can represent cell type split by laboratory, split by sample of origin :

plot_list = aquarius::plot_split_dimred(sobj,
                                        reduction = name2D,
                                        split_by = "laboratory",
                                        group_by = "cell_type",
                                        group_color = color_markers)

plot_list[[length(plot_list) + 1]] = patchwork::guide_area()

patchwork::wrap_plots(plot_list, nrow = 1) +
  patchwork::plot_layout(guides = "collect") &
  ggplot2::theme(legend.position = "right")

Location

We can represent cell type split by location, split by sample of origin :

plot_list = aquarius::plot_split_dimred(sobj,
                                        reduction = name2D,
                                        split_by = "location",
                                        group_by = "cell_type",
                                        group_color = color_markers)

plot_list[[length(plot_list) + 1]] = patchwork::guide_area()

patchwork::wrap_plots(plot_list, nrow = 1) +
  patchwork::plot_layout(guides = "collect") &
  ggplot2::theme(legend.position = "right")

Clusters

We can represent clusters, split by sample of origin :

plot_list = aquarius::plot_split_dimred(sobj,
                                        reduction = name2D,
                                        split_by = "project_name",
                                        group_by = "seurat_clusters",
                                        split_color = setNames(sample_info$color,
                                                               nm = sample_info$project_name),
                                        group_color = aquarius::gg_color_hue(length(levels(sobj$seurat_clusters))))

plot_list[[length(plot_list) + 1]] = clusters_plot +
  ggplot2::labs(title = "Cluster ID") &
  ggplot2::theme(plot.title = element_text(hjust = 0.5, size = 15))

patchwork::wrap_plots(plot_list, ncol = 4) &
  Seurat::NoLegend()

Cell cycle

We visualize cell cycle annotation, and BIRC5 and TOP2A expression levels :

plot_list = list()

# Seurat
plot_list[[1]] = Seurat::DimPlot(sobj, group.by = "Seurat.Phase",
                                 reduction = name2D) +
  Seurat::NoAxes() + ggplot2::labs(title = "Seurat annotation") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))

# cyclone
plot_list[[2]] = Seurat::DimPlot(sobj, group.by = "cyclone.Phase",
                                 reduction = name2D) +
  Seurat::NoAxes() + ggplot2::labs(title = "cyclone annotation") +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))

# BIRC5
plot_list[[3]] = Seurat::FeaturePlot(sobj, features = "BIRC5",
                                     reduction = name2D) +
  ggplot2::scale_color_gradientn(colors = aquarius::color_gene) +
  Seurat::NoAxes() +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))

# TK1
plot_list[[4]] = Seurat::FeaturePlot(sobj, features = "TOP2A",
                                     reduction = name2D) +
  ggplot2::scale_color_gradientn(colors = aquarius::color_gene) +
  Seurat::NoAxes() +
  ggplot2::theme(aspect.ratio = 1,
                 plot.title = element_text(hjust = 0.5))

patchwork::wrap_plots(plot_list, ncol = 2)

Save

We save the Seurat object :

saveRDS(sobj, file = paste0(out_dir, "/", save_name, "_sobj.rds"))

R Session

show

## R version 3.6.3 (2020-02-29)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.6 LTS
## 
## Matrix products: default
## BLAS:   /usr/local/lib/R/lib/libRblas.so
## LAPACK: /usr/local/lib/R/lib/libRlapack.so
## 
## locale:
## [1] C
## 
## attached base packages:
## [1] grid      stats     graphics  grDevices utils     datasets  methods  
## [8] base     
## 
## other attached packages:
## [1] ComplexHeatmap_2.14.0 ggplot2_3.3.5         patchwork_1.1.2      
## [4] dplyr_1.0.7          
## 
## loaded via a namespace (and not attached):
##   [1] softImpute_1.4              graphlayouts_0.7.0         
##   [3] pbapply_1.4-2               lattice_0.20-41            
##   [5] haven_2.3.1                 vctrs_0.3.8                
##   [7] usethis_2.0.1               dynwrap_1.2.1              
##   [9] blob_1.2.1                  survival_3.2-13            
##  [11] prodlim_2019.11.13          dynutils_1.0.5             
##  [13] later_1.3.0                 DBI_1.1.1                  
##  [15] R.utils_2.11.0              SingleCellExperiment_1.8.0 
##  [17] rappdirs_0.3.3              uwot_0.1.8                 
##  [19] dqrng_0.2.1                 jpeg_0.1-8.1               
##  [21] zlibbioc_1.32.0             pspline_1.0-18             
##  [23] pcaMethods_1.78.0           mvtnorm_1.1-1              
##  [25] htmlwidgets_1.5.4           GlobalOptions_0.1.2        
##  [27] future_1.22.1               UpSetR_1.4.0               
##  [29] laeken_0.5.2                leiden_0.3.3               
##  [31] clustree_0.4.3              parallel_3.6.3             
##  [33] scater_1.14.6               irlba_2.3.3                
##  [35] DEoptimR_1.0-9              tidygraph_1.1.2            
##  [37] Rcpp_1.0.9                  readr_2.0.2                
##  [39] KernSmooth_2.23-17          carrier_0.1.0              
##  [41] promises_1.1.0              gdata_2.18.0               
##  [43] DelayedArray_0.12.3         limma_3.42.2               
##  [45] graph_1.64.0                RcppParallel_5.1.4         
##  [47] Hmisc_4.4-0                 fs_1.5.2                   
##  [49] RSpectra_0.16-0             fastmatch_1.1-0            
##  [51] ranger_0.12.1               digest_0.6.25              
##  [53] png_0.1-7                   sctransform_0.2.1          
##  [55] cowplot_1.0.0               DOSE_3.12.0                
##  [57] ggvenn_0.1.9                here_1.0.1                 
##  [59] TInGa_0.0.0.9000            ggraph_2.0.3               
##  [61] pkgconfig_2.0.3             GO.db_3.10.0               
##  [63] DelayedMatrixStats_1.8.0    gower_0.2.1                
##  [65] ggbeeswarm_0.6.0            iterators_1.0.12           
##  [67] DropletUtils_1.6.1          reticulate_1.26            
##  [69] clusterProfiler_3.14.3      SummarizedExperiment_1.16.1
##  [71] circlize_0.4.15             beeswarm_0.4.0             
##  [73] GetoptLong_1.0.5            xfun_0.35                  
##  [75] bslib_0.3.1                 zoo_1.8-10                 
##  [77] tidyselect_1.1.0            reshape2_1.4.4             
##  [79] purrr_0.3.4                 ica_1.0-2                  
##  [81] pcaPP_1.9-73                viridisLite_0.3.0          
##  [83] rtracklayer_1.46.0          rlang_1.0.2                
##  [85] hexbin_1.28.1               jquerylib_0.1.4            
##  [87] dyneval_0.9.9               glue_1.4.2                 
##  [89] RColorBrewer_1.1-2          matrixStats_0.56.0         
##  [91] stringr_1.4.0               lava_1.6.7                 
##  [93] europepmc_0.3               DESeq2_1.26.0              
##  [95] recipes_0.1.17              labeling_0.3               
##  [97] harmony_0.1.0               httpuv_1.5.2               
##  [99] class_7.3-17                BiocNeighbors_1.4.2        
## [101] DO.db_2.9                   annotate_1.64.0            
## [103] jsonlite_1.7.2              XVector_0.26.0             
## [105] bit_4.0.4                   mime_0.9                   
## [107] aquarius_0.1.5              Rsamtools_2.2.3            
## [109] gridExtra_2.3               gplots_3.0.3               
## [111] stringi_1.4.6               processx_3.5.2             
## [113] gsl_2.1-6                   bitops_1.0-6               
## [115] cli_3.0.1                   batchelor_1.2.4            
## [117] RSQLite_2.2.0               randomForest_4.6-14        
## [119] tidyr_1.1.4                 data.table_1.14.2          
## [121] rstudioapi_0.13             org.Mm.eg.db_3.10.0        
## [123] GenomicAlignments_1.22.1    nlme_3.1-147               
## [125] qvalue_2.18.0               scran_1.14.6               
## [127] locfit_1.5-9.4              scDblFinder_1.1.8          
## [129] listenv_0.8.0               ggthemes_4.2.4             
## [131] gridGraphics_0.5-0          R.oo_1.24.0                
## [133] dbplyr_1.4.4                BiocGenerics_0.32.0        
## [135] TTR_0.24.2                  readxl_1.3.1               
## [137] lifecycle_1.0.1             timeDate_3043.102          
## [139] ggpattern_0.3.1             munsell_0.5.0              
## [141] cellranger_1.1.0            R.methodsS3_1.8.1          
## [143] proxyC_0.1.5                visNetwork_2.0.9           
## [145] caTools_1.18.0              codetools_0.2-16           
## [147] Biobase_2.46.0              GenomeInfoDb_1.22.1        
## [149] vipor_0.4.5                 lmtest_0.9-38              
## [151] msigdbr_7.5.1               htmlTable_1.13.3           
## [153] triebeard_0.3.0             lsei_1.2-0                 
## [155] xtable_1.8-4                ROCR_1.0-7                 
## [157] BiocManager_1.30.10         scatterplot3d_0.3-41       
## [159] abind_1.4-5                 farver_2.0.3               
## [161] parallelly_1.28.1           RANN_2.6.1                 
## [163] askpass_1.1                 GenomicRanges_1.38.0       
## [165] RcppAnnoy_0.0.16            tibble_3.1.5               
## [167] ggdendro_0.1-20             cluster_2.1.0              
## [169] future.apply_1.5.0          Seurat_3.1.5               
## [171] dendextend_1.15.1           Matrix_1.3-2               
## [173] ellipsis_0.3.2              prettyunits_1.1.1          
## [175] lubridate_1.7.9             ggridges_0.5.2             
## [177] igraph_1.2.5                RcppEigen_0.3.3.7.0        
## [179] fgsea_1.12.0                remotes_2.4.2              
## [181] scBFA_1.0.0                 destiny_3.0.1              
## [183] VIM_6.1.1                   testthat_3.1.0             
## [185] htmltools_0.5.2             BiocFileCache_1.10.2       
## [187] yaml_2.2.1                  utf8_1.1.4                 
## [189] plotly_4.9.2.1              XML_3.99-0.3               
## [191] ModelMetrics_1.2.2.2        e1071_1.7-3                
## [193] foreign_0.8-76              withr_2.5.0                
## [195] fitdistrplus_1.0-14         BiocParallel_1.20.1        
## [197] xgboost_1.4.1.1             bit64_4.0.5                
## [199] foreach_1.5.0               robustbase_0.93-9          
## [201] Biostrings_2.54.0           GOSemSim_2.13.1            
## [203] rsvd_1.0.3                  memoise_2.0.0              
## [205] evaluate_0.18               forcats_0.5.0              
## [207] rio_0.5.16                  geneplotter_1.64.0         
## [209] tzdb_0.1.2                  caret_6.0-86               
## [211] ps_1.6.0                    DiagrammeR_1.0.6.1         
## [213] curl_4.3                    fdrtool_1.2.15             
## [215] fansi_0.4.1                 highr_0.8                  
## [217] urltools_1.7.3              xts_0.12.1                 
## [219] GSEABase_1.48.0             acepack_1.4.1              
## [221] edgeR_3.28.1                checkmate_2.0.0            
## [223] scds_1.2.0                  cachem_1.0.6               
## [225] npsurv_0.4-0                babelgene_22.3             
## [227] rjson_0.2.20                openxlsx_4.1.5             
## [229] ggrepel_0.9.1               clue_0.3-60                
## [231] rprojroot_2.0.2             stabledist_0.7-1           
## [233] tools_3.6.3                 sass_0.4.0                 
## [235] nichenetr_1.1.1             magrittr_2.0.1             
## [237] RCurl_1.98-1.2              proxy_0.4-24               
## [239] car_3.0-11                  ape_5.3                    
## [241] ggplotify_0.0.5             xml2_1.3.2                 
## [243] httr_1.4.2                  assertthat_0.2.1           
## [245] rmarkdown_2.18              boot_1.3-25                
## [247] globals_0.14.0              R6_2.4.1                   
## [249] Rhdf5lib_1.8.0              nnet_7.3-14                
## [251] RcppHNSW_0.2.0              progress_1.2.2             
## [253] genefilter_1.68.0           statmod_1.4.34             
## [255] gtools_3.8.2                shape_1.4.6                
## [257] HDF5Array_1.14.4            BiocSingular_1.2.2         
## [259] rhdf5_2.30.1                splines_3.6.3              
## [261] AUCell_1.8.0                carData_3.0-4              
## [263] colorspace_1.4-1            generics_0.1.0             
## [265] stats4_3.6.3                base64enc_0.1-3            
## [267] dynfeature_1.0.0            smoother_1.1               
## [269] gridtext_0.1.1              pillar_1.6.3               
## [271] tweenr_1.0.1                sp_1.4-1                   
## [273] ggplot.multistats_1.0.0     rvcheck_0.1.8              
## [275] GenomeInfoDbData_1.2.2      plyr_1.8.6                 
## [277] gtable_0.3.0                zip_2.2.0                  
## [279] knitr_1.41                  latticeExtra_0.6-29        
## [281] biomaRt_2.42.1              IRanges_2.20.2             
## [283] fastmap_1.1.0               ADGofTest_0.3              
## [285] copula_1.0-0                doParallel_1.0.15          
## [287] AnnotationDbi_1.48.0        vcd_1.4-8                  
## [289] babelwhale_1.0.1            openssl_1.4.1              
## [291] scales_1.1.1                backports_1.2.1            
## [293] S4Vectors_0.24.4            ipred_0.9-12               
## [295] enrichplot_1.6.1            hms_1.1.1                  
## [297] ggforce_0.3.1               Rtsne_0.15                 
## [299] shiny_1.7.1                 numDeriv_2016.8-1.1        
## [301] polyclip_1.10-0             lazyeval_0.2.2             
## [303] Formula_1.2-3               tsne_0.1-3                 
## [305] crayon_1.3.4                MASS_7.3-54                
## [307] pROC_1.16.2                 viridis_0.5.1              
## [309] dynparam_1.0.0              rpart_4.1-15               
## [311] zinbwave_1.8.0              compiler_3.6.3             
## [313] ggtext_0.1.0